Statistical fluctuation
The IEC 60846-1:2009 §8.7.4 evaluation of reading scatter — the coefficient of variation at each point, the stepped limit that depends on the lower limit of the measuring range, and the two-out-of-w acceptance rule.
Last reviewed
Radiation measurement is a counting process, so repeated readings of a constant field vary. At high dose rates the event rate is high enough that the variation is negligible; near the lower limit of the measuring range it can dominate the total error. This test establishes whether the variation is within an acceptance limit that tightens as dose rate rises above that lower limit.
Coefficient of variation
At each test point, compute the coefficient of variation: the sample standard deviation of the readings, using the n−1 divisor, divided by the mean indication after background subtraction.
It is a relative quantity, which is the whole point — 0,3 µSv/h of scatter is negligible at 100 µSv/h and fatal at 1 µSv/h.
Data source
You do not run a separate irradiation campaign. §8.7.1 evaluates fluctuation from the same repeated readings taken for linearity, at every one of its test points. If linearity has w points with n readings each, fluctuation has the same w and the same n.
The tool supports running it as an independent test with its own points, which is occasionally useful when you are investigating a specific instrument. For a normal evaluation, use the linked mode — it is what the standard describes and it halves your beam time.
Stepped acceptance limit
The acceptance limit for CV is not fixed. It steps with dose rate, expressed relative to H₀, the lower limit of the measuring range:
- at H = H₀, the limit is at its most permissive — 15 %;
- between H₀ and 11·H₀ it tightens linearly, following (16 − H/H₀) %;
- from 11·H₀ upward it settles at the instrument’s general limit.
The basis is physical. Down at the lower limit of the range the instrument is counting a handful of events, and the counting statistics set a floor on the variation that no design can remove. Two decades higher that constraint no longer applies.
H₀ is a required input, and the tool withholds the verdict without it. Without H₀ there is no limit to compare against, and a verdict computed from an assumed H₀ is not traceable. Take it from the instrument’s specification, in the same unit as your test points.
Acceptance rule
Not every point has to be inside the limit. The rule works on the set:
- of the w points, w−2 must have CV below c₁ × limit;
- the remaining two points must be below c₂ × limit, and those two must not be adjacent to each other.
The coefficients c₁ and c₂ depend on w and n — more points and more readings per point make the test statistically stronger, and the coefficients reflect that.
Two consequences follow:
At least three points are required. With fewer, “w−2” stops meaning anything and the rule cannot be applied. The tool reports this rather than returning a verdict.
The two lenient points must not be adjacent. Two adjacent points with high variation indicate a defect localised in one part of the range, which is the condition the rule is designed to detect. Two isolated points with high variation are consistent with random variation. The adjacency condition distinguishes the two cases, and it is easily missed when reading a results table unaided.
Failure patterns
Uniformly high variation
This normally indicates the reading interval was too short — the readings are correlated, which paradoxically shows up as unstable behaviour rather than as artificially low scatter when the integration is fighting a changing field. Check that the field had stabilised before the first reading.
Variation concentrated at low points
This is expected; the stepped limit already allows for it. If it still fails there, the instrument’s usable lower limit is genuinely higher than its specification claims, and that is a finding worth writing down.
Variation concentrated at high points
This is not expected. Suspect dead-time behaviour, or an unstable source-to-detector geometry that moved between readings.